Repository navigation
Kafka Connect: Update runtime distribution LICENSE and NOTICE - #18019
Conversation
f055eed to
b0651a2
Compare
b0651a2 to
7ac7301
Compare
7ac7301 to
51d7956
Compare
51d7956 to
87bd092
Compare
|
@rmoff I think one of my main issues with the way this is currently structured is that we're both fixing the non-hive distribution and stripping the hive dist out. We should probably just focus on fixing the non-hive without removing the other (we can remove the build dependency, but not the full build config). |
The Hive distribution's LICENSE and NOTICE do not account for the jars it bundles, so a default build should not produce it and releases should not ship it. Its build configuration stays so that it can still be built explicitly with hiveDistZip. The distribution plugin adds every distribution's archives to assemble through the archives configuration, so the Hive archives are removed from that configuration rather than from the explicit assemble dependencies, which the plugin's wiring made redundant. Generated-by: Claude Code (Claude Fable 5.1) Co-Authored-By: Claude Fable 5.1 <[email protected]>
Replace the hand-picked AWS, GCP and Azure SDK dependencies with copies of the dependencies declared by the iceberg-aws-bundle, iceberg-gcp-bundle and iceberg-azure-bundle projects, taken from each bundle's implementation configuration once that project is evaluated. The distribution then offers the same features as the bundles, its LICENSE and NOTICE content can be kept in sync with theirs, and the lists cannot drift apart: the hand-picked list had already kept the AWS SDK's Apache HttpClient 4 module after the bundle and iceberg-aws moved to the HttpClient 5 module. The copied dependencies are resolved in this project, so its excludes, forced versions and conflict resolution apply, and the result is loose jars. The bundle jars themselves are not shipped: they would duplicate every class already present as a loose jar, and depending on them alone leaves unrelocated libraries such as Woodstox and commons-codec at versions that collide with the ones Hadoop needs. Also excludes the log4j binding, as iceberg-aws-bundle does, and keeps iceberg-bigquery, which the BigQuery Metastore catalog needs. Approach suggested by Dan Weeks in review. Generated-by: Claude Code (Claude Fable 5.1) Co-Authored-By: Claude Fable 5.1 <[email protected]>
…lists Generated with generateRuntimeDeps after switching the runtime to the cloud bundles' dependency lists. Generated-by: Claude Code (Claude Fable 5.1) Co-Authored-By: Claude Fable 5.1 <[email protected]>
Accounts for the third parties bundled in the non-Hive runtime distribution, checked jar by jar against the built zip, including code shaded inside other jars. Removes the jsr305 entry (now excluded) and stale entries for jars that are no longer bundled (Nimbus, json-smart, ASM, JCIP annotations, commons-beanutils). Adds entries for third parties that were bundled but unaccounted for, with full licence text for the non-Apache ones: Bouncy Castle (MIT); the MSV, RELAX NG Datatype (BSD) and isorelax (CDDL) components shaded in Woodstox; FastDoubleParser and Schubfach (MIT) shaded in Jackson; the Mozilla Public Suffix List (MPL 2.0) shipped by HttpComponents and Guava; BoringSSL and Tomcat Native linked into grpc-netty-shaded and Conscrypt; the BSD-licensed Bloom filter code in Hadoop and the derivative works in Commons Math; and the Apache-licensed code shaded in Parquet (Thrift, fastutil), gRPC xDS (Envoy, CEL, xDS, UDPA, PGV), OpenTelemetry, Reactor Netty and Azure SDK. Corrects entries that named the wrong licence: Google API Common, GAX and Auth Library are BSD 3-Clause; Jakarta Activation is EDL 1.0; JTS is listed under its EDL 1.0 option as in the Spark and Flink runtimes. Restores the copyright line in the ThreeTen Extra text, drops a duplicate JAXB entry, and adds Iceberg's own derived-code entries (Parquet, Kite, Presto, Delta Lake, Apache Commons, HttpComponents Client) as the Spark runtime LICENSE does. Cloud entries are taken from the aws/gcp/azure bundle LICENSE files; where those files were wrong (RELAX NG carried the Jakarta Activation text; the Google libraries were correct there but not here) the text is taken from the bundled jar or the upstream project instead. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Co-Authored-By: Claude Fable 5.1 <[email protected]> (cherry picked from commit 34c1944)
Carries the required third-party attributions for the bundled jars: Iceberg's own Kite, Presto and Delta Lake attributions, as in the NOTICE of every bundled iceberg-* jar and in the Spark and Flink runtimes; the cloud notices from the aws/gcp/azure bundle NOTICE files; and the substantive non-cloud notices (Apache Hadoop export-control, Apache Commons Math, Jakarta Activation, grpc-netty-shaded, Jackson including its Schubfach attribution, and Airlift Aircompressor as in the Spark runtime). The AWS SDK block reproduces the NOTICE shipped in the 2.54 jars in full. Boilerplate-only Apache notices are omitted as covered by the top-level Apache Iceberg attribution. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Co-Authored-By: Claude Fable 5.1 <[email protected]> (cherry picked from commit e892196)
…efault The default build now produces only the distribution without the Hive Metastore client. Document how to build the Hive distribution explicitly and drop the suggestion to pick between two distributions. Generated-by: Claude Code (Claude Fable 5.1) Co-Authored-By: Claude Fable 5.1 <[email protected]>
The libraries-bom 26.88 bump (apache#18094) adds the dev.cel:common, dev.cel:protobuf and dev.cel:runtime jars to the distribution. Add the same CEL-Java entry that apache#18226 added to gcp-bundle/LICENSE, and note that the CEL specification is also bundled by CEL-Java. Co-Authored-By: Claude Fable 5.1 <[email protected]> (cherry picked from commit 87bd092)
87bd092 to
4ffcf4a
Compare
List the Presto and Kite code the bundled jars contain, drop the Spark-only Delta Lake item, and add an entry for the APSL-licensed dnsinfo.h header compiled into netty-resolver-dns-native-macos. Generated-by: Claude Code (Claude Fable 5.1) Co-Authored-By: Claude Fable 5.1 <[email protected]>
|
Thanks @rmoff! |
…#18019) * Kafka Connect: Update runtime distribution LICENSE and NOTICE Co-authored-by: Ryan Blue <[email protected]> Co-authored-by: Claude Fable 5.1 <[email protected]>
|
Looks like this is only for the I'll take a look at adding the same for |
Thanks @kevinjqliu. Yes this was just for the |
The Hive distribution's LICENSE and NOTICE predate apache#18019 and no longer match what it bundles. The Hive distribution is now the main one plus the Hive Metastore client and its dependencies, so its LICENSE and NOTICE are the main ones followed by entries for those additions. The added entries cover the Hive Metastore client and its Hadoop, Thrift, ZooKeeper, DataNucleus, Derby, Jetty, Tomcat and logging dependencies, plus code bundled inside them, such as ASM, FDLIBM, quiche and the YARN web UI's jQuery, DataTables and jsTree. Generated-by: GitHub Copilot (Claude Opus 5.5) Co-authored-by: Copilot App <[email protected]>
#15212 is blocked on the Kafka Connect runtime distribution's
LICENSEandNOTICE. This PR reconciles both against the bundled jars and regeneratesruntime-deps.txtsocheckRuntimeDepsguards the inventory. The cloud SDK dependencies are copied from theiceberg-aws-bundle,iceberg-gcp-bundleandiceberg-azure-bundledeclarations rather than listed by hand. The Hive distribution is not reconciled, soassembleno longer produces it; its build configuration stays. Continues @rdblue's #16147; the jsr305 commit is his.Test plan:
./gradlew :iceberg-kafka-connect:iceberg-kafka-connect-runtime:checkpasses (checkRuntimeDeps,checkClassUniqueness, 14 integration tests against the installed distribution);assemble --dry-runschedules only the main distribution andhiveDistZipstill builds on request. Every jar in the zip was opened to list shaded packages, native libraries and embedded licence files, and each component's licence read from its POM or embedded text.AI Disclosure